From nicolas.brodu at numerimoire.net Wed Feb 4 23:33:47 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Thu, 5 Feb 2009 00:33:47 +0100 Subject: [Causality-ML] Reminder: Conditional independence with positive definite kernels Message-ID: <200902050033.47491.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation by Kenji Fukumizu. Topic: Conditional independence with positive definite kernels When: Friday 06/02, Tokyo 8h, Paris 0h, ET 18h (Thursday), PT 15h (Thursday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090206KF Phone number: +1 (218) 936-7999 Participant code: 665140# Kenji's talk builds on and extends the last presentation by Arthur Gretton. You may wish to view the video replay of the last presentation at: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090129AG Abstract of tomorrow's presentation: -------- A new nonparametric methodology for dependence of random variables is discussed. It uses the framework of reproducing kernel Hilbert spaces (RKHS) defined by positive definite kernels. In this methodology, a random variable is mapped to a RKHS, thus random variables on the RKHS are considered. The basic statistics such as mean and covariance of the variables on the RKHS can capture all the information on the underlying probabilities by using a sufficiently rich function space as RKHS, which we call a characteristic RKHS. In this presentation, I focus on the characterization of conditional independence of variables by covariance operators. I will show that, in the similar fashion to the partial correlation for Gaussian variables, the conditional covariance operator on characteristic RKHS characterizes the conditional independence of arbitrary random variables. The Hilbert-Schmidt norm of the conditional covariance operator, thus, defines a measure of conditional dependence, and the empirical estimate of the norm works as a test statistics of conditional independence. In addition, I will show that, for any characteristic RKHS the Hilbert-Schmidt norm of the "normalized" conditional covariance operator coincides with the (conditional) mean square contingency in population, and thus does not depends on the choice of kernel. The empirical estimate gives a new kernel estimate of the mean square contingency. I will also show consistency of the estimators, and demonstrate some experimental results of conditional independence tests with the Hilbert-Schmidt norm of the normalized and unnormalized conditional covariance operators. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From nicolas.brodu at numerimoire.net Mon Feb 9 09:00:14 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Mon, 9 Feb 2009 10:00:14 +0100 Subject: [Causality-ML] Video replay and next talks announcement Message-ID: <200902091000.15009.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, The video of the last talk by Kenji Fukumizu is available online! You can find it together with other videos for most of the recent talks on the schedule and material page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture The next presentation is by Dominik Janzing. He will present us with the task of distinguishing between cause and effect. When: Thursday 12 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090212DJ This presentation will be followed by a presentation by Kun Zhang on a method to solve this problem: When: Thursday 19 February 2009, time TBA URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ The following week, on a different topic, I will present you how to define the states of a process that lead to the same decisions given a utility function, with reconstruction possible from data: When: Thursday 26 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090226NB You can find a schedule of planned future talks by following the first link of this message. If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From nicolas.brodu at numerimoire.net Wed Feb 11 17:06:53 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Wed, 11 Feb 2009 18:06:53 +0100 Subject: [Causality-ML] Reminder: Cause effect pairs: Distinguishing between cause and effect Message-ID: <200902111806.53261.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation by Joris Mooij and Dominik Janzing. Topic: Cause effect pairs: Distinguishing between cause and effect When: Thursday 12 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090212DJ Phone number: +1 (218) 936-7999 Participant code: 665140# Dominik and Joris will present us with the CauseEffectPairs task of the causality challenge that was the topic of the NIPS2008 workshop. Next week presentation by Kun Zhang will be on the method he applied to answer the challenge. A planning of future talks is available on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture Abstract of tomorrow's presentation: -------- We describe eight data sets that together formed the CauseEffectPairs task in the Causality Challenge #2: Pot-Luck competition. Each data set consists of a sample of a pair of statistically dependent random variables. For each data set, one variable is known to cause the other one, but this information was hidden from the participants; the task for the participants was to identify which of the two variables was the cause and which one the effect, based upon the observed sample. The data sets were chosen such that we expect common agreement on the ground truth. Even though part of the statistical dependences may also be due to hidden common causes, common sense tells us that there is a significant cause-effect relation between the two variables in each data set. We also present encouraging baseline results using a recently developed causal inference method that uses additive noise models. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From nicolas.brodu at numerimoire.net Fri Feb 13 14:58:25 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Fri, 13 Feb 2009 15:58:25 +0100 Subject: [Causality-ML] Video replay available, next talk announcement Message-ID: <200902131558.25408.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, The video of the last talk by Joris Mooij and Dominik Janzing is available online: http://www.encours.org/causality/causeEffectPairs/replay.html The next presentation is by Kun Zhang. He will present us how to distinguish causes from effects by using a non-linear acyclic causal model and Independent Component Analysis (ICA). When: Thursday 19 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ This presentation directly follows yesterday's one. You can find a schedule of future talks on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture Abstract -------- Distinguishing causes from effects is an important problem in many areas. In this talk, we present a very general but well defined nonlinear acyclic causal model, namely, post-nonlinear acyclic causal model with inner additive noise, to tackle this problem. In this model, each observed variable is generated by a nonlinear function of its parents, with additive noise, followed by a nonlinear transformation. The nonlinear transformation in the second stage takes into account the effect of sensor or measurement distortions, which are usually encountered in practice. In the two-variable case, if all the nonlinearities involved in the model are invertible, by relating the proposed model to the post-nonlinear independent component analysis (ICA) problem, we give the conditions under which the causal relation can be uniquely found. We present a two-step method, which is constrained nonlinear ICA followed by statistical independence tests, to examine if a given causal direction holds in the two-variable case. We apply this method to solve the task "CauseEffectPairs" proposed by Mooij, Janzing, and Sch?lkopf in the Pot-luck challenge, and successfully identify causes from effects for all eight problems. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From nicolas.brodu at numerimoire.net Wed Feb 18 15:33:01 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Wed, 18 Feb 2009 16:33:01 +0100 Subject: [Causality-ML] Reminder: Distinguishing Causes from Effects using Nonlinear Acyclic Causal Models Message-ID: <200902181633.02005.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation by Kun Zhang: Distinguishing Causes from Effects using Nonlinear Acyclic Causal Models Thursday 19 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ Phone number: +1 (218) 936-7999 Participant code: 665140# This presentation directly follows last week's one: Kun Zhang will present the method he applied to answer the challenge on cause-effect pairs. The video of last week presentation is available at: http://www.encours.org/causality/causeEffectPairs/replay.html Abstract of tomorrow's presentation: -------- Distinguishing causes from effects is an important problem in many areas. In this talk, we present a very general but well defined nonlinear acyclic causal model, namely, post-nonlinear acyclic causal model with inner additive noise, to tackle this problem. In this model, each observed variable is generated by a nonlinear function of its parents, with additive noise, followed by a nonlinear transformation. The nonlinear transformation in the second stage takes into account the effect of sensor or measurement distortions, which are usually encountered in practice. In the two-variable case, if all the nonlinearities involved in the model are invertible, by relating the proposed model to the post-nonlinear independent component analysis (ICA) problem, we give the conditions under which the causal relation can be uniquely found. We present a two-step method, which is constrained nonlinear ICA followed by statistical independence tests, to examine if a given causal direction holds in the two-variable case. We apply this method to solve the task "CauseEffectPairs" proposed by Mooij, Janzing, and Sch?lkopf in the Pot-luck challenge, and successfully identify causes from effects for all eight problems. -------- A planning of future talks is available on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From guyon at clopinet.com Thu Feb 19 07:04:07 2009 From: guyon at clopinet.com (Isabelle Guyon) Date: Wed, 18 Feb 2009 23:04:07 -0800 Subject: [Causality-ML] Reminder: Distinguishing Causes from Effects using Nonlinear Acyclic Causal Models References: <200902181633.02005.nicolas.brodu@numerimoire.net> Message-ID: <499D0467.5020202@clopinet.com> I am here but you do not hear me Nicolas Brodu wrote: >Dear Causality and Machine Learning group, > >This is a friendly reminder for tomorrow's presentation by Kun Zhang: > > Distinguishing Causes from Effects using Nonlinear Acyclic Causal Models > > Thursday 19 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) > http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ > Phone number: +1 (218) 936-7999 > Participant code: 665140# > >This presentation directly follows last week's one: Kun Zhang will present the >method he applied to answer the challenge on cause-effect pairs. The video of >last week presentation is available at: > http://www.encours.org/causality/causeEffectPairs/replay.html > >Abstract of tomorrow's presentation: >-------- >Distinguishing causes from effects is an important problem in many areas. In >this talk, we present a very general but well defined nonlinear acyclic >causal model, namely, post-nonlinear acyclic causal model with inner additive >noise, to tackle this problem. In this model, each observed variable is >generated by a nonlinear function of its parents, with additive noise, >followed by a nonlinear transformation. The nonlinear transformation in the >second stage takes into account the effect of sensor or measurement >distortions, which are usually encountered in practice. In the two-variable >case, if all the nonlinearities involved in the model are invertible, by >relating the proposed model to the post-nonlinear independent component >analysis (ICA) problem, we give the conditions under which the causal >relation can be uniquely found. We present a two-step method, which is >constrained nonlinear ICA followed by statistical independence tests, to >examine if a given causal direction holds in the two-variable case. We apply >this method to solve the task "CauseEffectPairs" proposed by Mooij, Janzing, >and Sch?lkopf in the Pot-luck challenge, and successfully identify causes >from effects for all eight problems. >-------- > >A planning of future talks is available on the group page: > http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture > >If you know of potentially interested speakers or if you wish to present a >paper, please send us a message so we can add you in the planning. > >Best regards, >Nicolas Brodu >_______________________________________________ >Causality and Machine Learning >Mailing-list subscription: >http://mail.encours.org/listinfo/causality-ml >Material and presentation schedule: >http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture > > From benmessaoud.montassar at hotmail.fr Thu Feb 19 16:16:24 2009 From: benmessaoud.montassar at hotmail.fr (benmessaoud.montassar at hotmail.fr) Date: Thu, 19 Feb 2009 08:16:24 -0800 Subject: [Causality-ML] =?iso-8859-1?q?R=E9ponse_en_cas_d=27absence?= In-Reply-To: <499D0467.5020202@clopinet.com> Message-ID: An HTML attachment was scrubbed... URL: From benmessaoud.montassar at hotmail.fr Thu Feb 19 16:17:01 2009 From: benmessaoud.montassar at hotmail.fr (benmessaoud.montassar at hotmail.fr) Date: Thu, 19 Feb 2009 08:17:01 -0800 Subject: [Causality-ML] =?iso-8859-1?q?R=E9ponse_en_cas_d=27absence?= In-Reply-To: Message-ID: An HTML attachment was scrubbed... URL: From subramani.mani at Vanderbilt.Edu Thu Feb 19 16:17:43 2009 From: subramani.mani at Vanderbilt.Edu (Mani, Subramani) Date: Thu, 19 Feb 2009 10:17:43 -0600 Subject: [Causality-ML] Causality-ML Digest, Vol 1, Issue 1 In-Reply-To: References: Message-ID: Hi there I can hear you. - subramani ---------------------------------------------------------------------- Subramani Mani Department of Biomedical Informatics Vanderbilt University 400 Eskind Biomedical Library 2209 Garland Avenue Nashville, TN 37232-8340 Email: subramani.mani at vanderbilt.edu Web: http://discover.mc.vanderbilt.edu/manis/ Phone: 615-936-2880 Fax: 615-936-1427 ---------------------------------------------------------------------- -----Original Message----- From: causality-ml-bounces at encours.org [mailto:causality-ml-bounces at encours.org] On Behalf Of causality-ml-request at encours.org Sent: Friday, February 13, 2009 8:58 AM To: causality-ml at encours.org Subject: Causality-ML Digest, Vol 1, Issue 1 Send Causality-ML mailing list submissions to causality-ml at encours.org To subscribe or unsubscribe via the World Wide Web, visit http://mail.encours.org/listinfo/causality-ml or, via email, send a message with subject or body 'help' to causality-ml-request at encours.org You can reach the person managing the list at causality-ml-owner at encours.org When replying, please edit your Subject line so it is more specific than "Re: Contents of Causality-ML digest..." Today's Topics: 1. Test (Nicolas Brodu) 2. test 2 (Nicolas Brodu) 3. Test 4 (Nicolas Brodu) 4. test from free (nicolas.brodu at free.fr) 5. test from free (nicolas.brodu at free.fr) 6. another test (Nicolas Brodu) 7. Test 3 (Nicolas Brodu) 8. Test (Nicolas Brodu) 9. Test 3 (Nicolas Brodu) 10. Video replay, mailing list and next talk announcement (Nicolas Brodu) 11. Reminder: Conditional independence with positive definite kernels (Nicolas Brodu) 12. Video replay and next talks announcement (Nicolas Brodu) 13. Reminder: Cause effect pairs: Distinguishing between cause and effect (Nicolas Brodu) 14. Video replay available, next talk announcement (Nicolas Brodu) ---------------------------------------------------------------------- Message: 1 Date: Fri, 30 Jan 2009 17:57:40 +0100 From: Nicolas Brodu Subject: [Causality-ML] Test To: causality-ml at encours.org Message-ID: <200901301757.40589.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="iso-8859-1" post ? ------------------------------ Message: 2 Date: Fri, 30 Jan 2009 18:00:12 +0100 From: Nicolas Brodu Subject: [Causality-ML] test 2 To: causality-ml at encours.org Message-ID: <200901301800.13045.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="us-ascii" ha? ------------------------------ Message: 3 Date: Fri, 30 Jan 2009 20:06:47 +0100 From: Nicolas Brodu Subject: [Causality-ML] Test 4 To: causality-ml at encours.org Message-ID: <200901302006.47697.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" via fai ------------------------------ Message: 4 Date: Fri, 30 Jan 2009 20:07:51 +0100 From: nicolas.brodu at free.fr Subject: [Causality-ML] test from free To: causality-ml at encours.org Message-ID: <20090130200751.zyr3ll59ssooc044 at imp4.free.fr> Content-Type: text/plain; charset=ISO-8859-1; DelSp="Yes"; format="flowed" haha! ------------------------------ Message: 5 Date: Fri, 30 Jan 2009 18:58:07 +0100 From: nicolas.brodu at free.fr Subject: [Causality-ML] test from free To: causality-ml at encours.org Message-ID: <20090130185807.mo7indqsgkssc84k at imp4.free.fr> Content-Type: text/plain; charset=ISO-8859-1; DelSp="Yes"; format="flowed" another FAI ------------------------------ Message: 6 Date: Fri, 30 Jan 2009 20:19:20 +0100 From: Nicolas Brodu Subject: [Causality-ML] another test To: causality-ml at encours.org Message-ID: <200901302019.20910.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" trntr sr ------------------------------ Message: 7 Date: Fri, 30 Jan 2009 18:00:57 +0100 From: Nicolas Brodu Subject: [Causality-ML] Test 3 To: causality-ml at encours.org Message-ID: <200901301800.58013.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" via fai ------------------------------ Message: 8 Date: Fri, 30 Jan 2009 17:39:54 +0100 From: Nicolas Brodu Subject: [Causality-ML] Test To: causality-ml at encours.org Message-ID: <200901301739.54711.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="iso-8859-1" post ? ------------------------------ Message: 9 Date: Fri, 30 Jan 2009 18:55:59 +0100 From: Nicolas Brodu Subject: [Causality-ML] Test 3 To: causality-ml at encours.org Message-ID: <200901301855.59780.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" via fai ------------------------------ Message: 10 Date: Fri, 30 Jan 2009 22:53:30 +0100 From: Nicolas Brodu Subject: [Causality-ML] Video replay, mailing list and next talk announcement To: Causality and Machine Learning Message-ID: <200901302253.30168.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" Dear Causality and Machine Learning group, We have a new mailing-list! Please do not hesitate to use it and discuss the papers and presentations. Future events and talks will be announced to the list as well. You can manage your options (ex: changing address, unsubscribing duplicates...) by following the links sent to you in the list welcome message. The video of the last talk by Arthur Gretton is available online! You can find it together with other videos for most of the recent talks on the schedule and material page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture The next presentation is by Kenji Fukumizu. It builds on and extends the last presentation: Topic: Conditional independence with positive definite kernels When: Friday 06/02, Tokyo 8h, Paris 0h, ET 18h (Thursday), PT 15h (Thursday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090206KF Abstract: -------- A new nonparametric methodology for dependence of random variables is discussed. It uses the framework of reproducing kernel Hilbert spaces (RKHS) defined by positive definite kernels. In this methodology, a random variable is mapped to a RKHS, thus random variables on the RKHS are considered. The basic statistics such as mean and covariance of the variables on the RKHS can capture all the information on the underlying probabilities by using a sufficiently rich function space as RKHS, which we call a characteristic RKHS. In this presentation, I focus on the characterization of conditional independence of variables by covariance operators. I will show that, in the similar fashion to the partial correlation for Gaussian variables, the conditional covariance operator on characteristic RKHS characterizes the conditional independence of arbitrary random variables. The Hilbert-Schmidt norm of the conditional covariance operator, thus, defines a measure of conditional dependence, and the empirical estimate of the norm works as a test statistics of conditional independence. In addition, I will show that, for any characteristic RKHS the Hilbert-Schmidt norm of the "normalized" conditional covariance operator coincides with the (conditional) mean square contingency in population, and thus does not depends on the choice of kernel. The empirical estimate gives a new kernel estimate of the mean square contingency. I will also show consistency of the estimators, and demonstrate some experimental results of conditional independence tests with the Hilbert-Schmidt norm of the normalized and unnormalized conditional covariance operators. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu ------------------------------ Message: 11 Date: Thu, 5 Feb 2009 00:33:47 +0100 From: Nicolas Brodu Subject: [Causality-ML] Reminder: Conditional independence with positive definite kernels To: Causality and Machine Learning Message-ID: <200902050033.47491.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation by Kenji Fukumizu. Topic: Conditional independence with positive definite kernels When: Friday 06/02, Tokyo 8h, Paris 0h, ET 18h (Thursday), PT 15h (Thursday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090206KF Phone number: +1 (218) 936-7999 Participant code: 665140# Kenji's talk builds on and extends the last presentation by Arthur Gretton. You may wish to view the video replay of the last presentation at: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090129AG Abstract of tomorrow's presentation: -------- A new nonparametric methodology for dependence of random variables is discussed. It uses the framework of reproducing kernel Hilbert spaces (RKHS) defined by positive definite kernels. In this methodology, a random variable is mapped to a RKHS, thus random variables on the RKHS are considered. The basic statistics such as mean and covariance of the variables on the RKHS can capture all the information on the underlying probabilities by using a sufficiently rich function space as RKHS, which we call a characteristic RKHS. In this presentation, I focus on the characterization of conditional independence of variables by covariance operators. I will show that, in the similar fashion to the partial correlation for Gaussian variables, the conditional covariance operator on characteristic RKHS characterizes the conditional independence of arbitrary random variables. The Hilbert-Schmidt norm of the conditional covariance operator, thus, defines a measure of conditional dependence, and the empirical estimate of the norm works as a test statistics of conditional independence. In addition, I will show that, for any characteristic RKHS the Hilbert-Schmidt norm of the "normalized" conditional covariance operator coincides with the (conditional) mean square contingency in population, and thus does not depends on the choice of kernel. The empirical estimate gives a new kernel estimate of the mean square contingency. I will also show consistency of the estimators, and demonstrate some experimental results of conditional independence tests with the Hilbert-Schmidt norm of the normalized and unnormalized conditional covariance operators. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu ------------------------------ Message: 12 Date: Mon, 9 Feb 2009 10:00:14 +0100 From: Nicolas Brodu Subject: [Causality-ML] Video replay and next talks announcement To: Causality and Machine Learning Message-ID: <200902091000.15009.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" Dear Causality and Machine Learning group, The video of the last talk by Kenji Fukumizu is available online! You can find it together with other videos for most of the recent talks on the schedule and material page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture The next presentation is by Dominik Janzing. He will present us with the task of distinguishing between cause and effect. When: Thursday 12 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090212DJ This presentation will be followed by a presentation by Kun Zhang on a method to solve this problem: When: Thursday 19 February 2009, time TBA URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ The following week, on a different topic, I will present you how to define the states of a process that lead to the same decisions given a utility function, with reconstruction possible from data: When: Thursday 26 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090226NB You can find a schedule of planned future talks by following the first link of this message. If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu ------------------------------ Message: 13 Date: Wed, 11 Feb 2009 18:06:53 +0100 From: Nicolas Brodu Subject: [Causality-ML] Reminder: Cause effect pairs: Distinguishing between cause and effect To: Causality and Machine Learning Message-ID: <200902111806.53261.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation by Joris Mooij and Dominik Janzing. Topic: Cause effect pairs: Distinguishing between cause and effect When: Thursday 12 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090212DJ Phone number: +1 (218) 936-7999 Participant code: 665140# Dominik and Joris will present us with the CauseEffectPairs task of the causality challenge that was the topic of the NIPS2008 workshop. Next week presentation by Kun Zhang will be on the method he applied to answer the challenge. A planning of future talks is available on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture Abstract of tomorrow's presentation: -------- We describe eight data sets that together formed the CauseEffectPairs task in the Causality Challenge #2: Pot-Luck competition. Each data set consists of a sample of a pair of statistically dependent random variables. For each data set, one variable is known to cause the other one, but this information was hidden from the participants; the task for the participants was to identify which of the two variables was the cause and which one the effect, based upon the observed sample. The data sets were chosen such that we expect common agreement on the ground truth. Even though part of the statistical dependences may also be due to hidden common causes, common sense tells us that there is a significant cause-effect relation between the two variables in each data set. We also present encouraging baseline results using a recently developed causal inference method that uses additive noise models. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu ------------------------------ Message: 14 Date: Fri, 13 Feb 2009 15:58:25 +0100 From: Nicolas Brodu Subject: [Causality-ML] Video replay available, next talk announcement To: Causality and Machine Learning Message-ID: <200902131558.25408.nicolas.brodu at numerimoire.net> Content-Type: text/plain; charset="utf-8" Dear Causality and Machine Learning group, The video of the last talk by Joris Mooij and Dominik Janzing is available online: http://www.encours.org/causality/causeEffectPairs/replay.html The next presentation is by Kun Zhang. He will present us how to distinguish causes from effects by using a non-linear acyclic causal model and Independent Component Analysis (ICA). When: Thursday 19 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090219KZ This presentation directly follows yesterday's one. You can find a schedule of future talks on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture Abstract -------- Distinguishing causes from effects is an important problem in many areas. In this talk, we present a very general but well defined nonlinear acyclic causal model, namely, post-nonlinear acyclic causal model with inner additive noise, to tackle this problem. In this model, each observed variable is generated by a nonlinear function of its parents, with additive noise, followed by a nonlinear transformation. The nonlinear transformation in the second stage takes into account the effect of sensor or measurement distortions, which are usually encountered in practice. In the two-variable case, if all the nonlinearities involved in the model are invertible, by relating the proposed model to the post-nonlinear independent component analysis (ICA) problem, we give the conditions under which the causal relation can be uniquely found. We present a two-step method, which is constrained nonlinear ICA followed by statistical independence tests, to examine if a given causal direction holds in the two-variable case. We apply this method to solve the task "CauseEffectPairs" proposed by Mooij, Janzing, and Sch?lkopf in the Pot-luck challenge, and successfully identify causes from effects for all eight problems. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu ------------------------------ _______________________________________________ Causality-ML mailing list Causality-ML at encours.org http://mail.encours.org/listinfo/causality-ml End of Causality-ML Digest, Vol 1, Issue 1 ****************************************** From benmessaoud.montassar at hotmail.fr Thu Feb 19 16:17:52 2009 From: benmessaoud.montassar at hotmail.fr (benmessaoud.montassar at hotmail.fr) Date: Thu, 19 Feb 2009 08:17:52 -0800 Subject: [Causality-ML] =?iso-8859-1?q?R=E9ponse_en_cas_d=27absence?= In-Reply-To: Message-ID: An HTML attachment was scrubbed... URL: From nicolas.brodu at numerimoire.net Thu Feb 19 16:27:35 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Thu, 19 Feb 2009 17:27:35 +0100 Subject: [Causality-ML] The presentation is on In-Reply-To: References: Message-ID: <200902191727.35172.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, There were some technical issues with the phone system, but it seems to be working again. The presentation has now started, please join in and ask questions to the lecturer! http://www.encours.org/causality/nlinAcyclicCausalModel/auditor.html Phone number: +1 (218) 936-7999 Participant code: 665140# Nicolas Thursday 19 February 2009 17:17:43 Mani Subramani wrote: > Hi there > > I can hear you. > > - subramani > > ---------------------------------------------------------------------- > Subramani Mani > Department of Biomedical Informatics > Vanderbilt University > 400 Eskind Biomedical Library > 2209 Garland Avenue > Nashville, TN 37232-8340 > Email: subramani.mani at vanderbilt.edu > Web: http://discover.mc.vanderbilt.edu/manis/ > Phone: 615-936-2880 > Fax: 615-936-1427 > ---------------------------------------------------------------------- From nicolas.brodu at numerimoire.net Fri Feb 20 09:51:38 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Fri, 20 Feb 2009 10:51:38 +0100 Subject: [Causality-ML] Video replay available, next talk announcement Message-ID: <200902201051.39269.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, The video of the last talk by Kun Zhang is available online: http://www.encours.org/causality/nlinAcyclicCausalModel/replay.html The next presentation is by Nicolas Brodu. I will present how to infer the states of a process that lead to the same decisions given a utility function. When: Thursday 26 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090226NB We are sorry for the technical difficulties some of us had yesterday and for the few messages of spam on the list. Actions have been taken in order to improve these points. Abstract of the next presentation: -------- We are monitoring a system, and we are given a utility/cost function for comparing predictions made about this system to what happens really. We would like to get: ? The optimal decisions to take at any given time, in order to maximise the utility function. ? The corresponding expected utility. ? Events when one of the above quantity change in the system. In this presentation I link the above questions to the causal states of the system: sets of observable with the same conditional distribution of predictions. The causal states define the inherent structure of the system, they form a Markovian graph called the ?-machine. The utility function then encodes some a priori knowledge representing the interests of the observer. The decisional states are the consequence of the utility function: they form a sub-machine of the ?-machine, whose structure not only reflects the inherent relations present in the data but also considers the information introduced by the utility function. The transitions between these decisional states correspond to events that lead to a change of decision. Example applications are given for discrete process hidden state reconstruction and image filtering. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From nicolas.brodu at numerimoire.net Thu Feb 26 01:29:18 2009 From: nicolas.brodu at numerimoire.net (Nicolas Brodu) Date: Thu, 26 Feb 2009 02:29:18 +0100 Subject: [Causality-ML] Decisional States (Reminder) + Live audio streaming! Message-ID: <200902260229.19617.nicolas.brodu@numerimoire.net> Dear Causality and Machine Learning group, This is a friendly reminder for tomorrow's presentation on Decisional States, by Nicolas Brodu When: Thursday 26 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090226NB NEW FEATURE: You can listen to the audio from the Internet! Just open the presentation link at the indicated time and hope for the best (the system is still experimental). You can also ask questions directly from your web browser. Please phone the usual number to participate to the discussion: Phone number: +1 (218) 936-7999 Participant code: 665140# A planning of future talks is available on the group page: http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture Abstract of tomorrow's presentation: -------- We are monitoring a system, and we are given a utility/cost function for comparing predictions made about this system to what happens really. We would like to get: ? The optimal decisions to take at any given time, in order to maximise the utility function. ? The corresponding expected utility. ? Events when one of the above quantity change in the system. In this presentation I link the above questions to the causal states of the system: sets of observable with the same conditional distribution of predictions. The causal states define the inherent structure of the system, they form a Markovian graph called the ?-machine. The utility function then encodes some a priori knowledge representing the interests of the observer. The decisional states are the consequence of the utility function: they form a sub-machine of the ?-machine, whose structure not only reflects the inherent relations present in the data but also considers the information introduced by the utility function. The transitions between these decisional states correspond to events that lead to a change of decision. Example applications are given for discrete process hidden state reconstruction and image filtering. -------- If you know of potentially interested speakers or if you wish to present a paper, please send us a message so we can add you in the planning. Best regards, Nicolas Brodu From guyon at clopinet.com Thu Feb 26 07:42:03 2009 From: guyon at clopinet.com (Isabelle Guyon) Date: Wed, 25 Feb 2009 23:42:03 -0800 Subject: [Causality-ML] Decisional States (Reminder) + Live audio streaming! References: <200902260229.19617.nicolas.brodu@numerimoire.net> Message-ID: <49A647CB.7050300@clopinet.com> on a ete coupes... Nicolas Brodu wrote: >Dear Causality and Machine Learning group, > >This is a friendly reminder for tomorrow's presentation on Decisional States, >by Nicolas Brodu > > When: Thursday 26 February 2009, Paris 17h, ET 11h, PT 8h, Tokyo 1h(Friday) > URL: http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture090226NB > >NEW FEATURE: You can listen to the audio from the Internet! > >Just open the presentation link at the indicated time and hope for the best >(the system is still experimental). You can also ask questions directly from >your web browser. > >Please phone the usual number to participate to the discussion: > Phone number: +1 (218) 936-7999 > Participant code: 665140# > >A planning of future talks is available on the group page: > http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture > >Abstract of tomorrow's presentation: >-------- >We are monitoring a system, and we are given a utility/cost function for >comparing predictions made about this system to what happens really. We would >like to get: > ? The optimal decisions to take at any given time, in order to maximise the >utility function. > ? The corresponding expected utility. > ? Events when one of the above quantity change in the system. > In this presentation I link the above questions to the causal states of the >system: sets of observable with the same conditional distribution of >predictions. The causal states define the inherent structure of the system, >they form a Markovian graph called the ?-machine. The utility function then >encodes some a priori knowledge representing the interests of the observer. >The decisional states are the consequence of the utility function: they form >a sub-machine of the ?-machine, whose structure not only reflects the >inherent relations present in the data but also considers the information >introduced by the utility function. The transitions between these decisional >states correspond to events that lead to a change of decision. Example >applications are given for discrete process hidden state reconstruction and >image filtering. >-------- > >If you know of potentially interested speakers or if you wish to present a >paper, please send us a message so we can add you in the planning. > >Best regards, >Nicolas Brodu >_______________________________________________ >Causality and Machine Learning >Mailing-list subscription: >http://mail.encours.org/listinfo/causality-ml >Material and presentation schedule: >http://www.afia-france.org/tiki-index.php?page=Groupe+de+lecture > >